Impact of Generative Artificial Intelligence on Footwear Design Concept and Ideation ^†

Abstract

The impact of Generative artificial intelligence (GAI) on footwear design creativity and feasibility was investigated in this study. Using a text-to-image GAI tool called Midjourney, 17 prompts were tested to generate footwear concepts. Ten distinct outcomes were selected from the results and evaluated by seven experts in footwear design. Prompt implementation correlated weakly positively with design creativity and feasibility. A set of prompts in the Japanese style showed significantly higher creativity due to explicit style descriptions. Sharp generative design features had higher creativity but lower feasibility. Concepts generated without subcategories induced lower feasibility and prompt implementation. These findings offer insights into GAI’s role in footwear design innovation.

1. Introduction

Artificial intelligence (AI) has been rapidly adopted by the industry, revolutionizing various aspects of our lives. Generative artificial intelligence (GAI) has especially shown extensive and profound advancements and significantly impacted the global economy [1]. GAI generates new content of text, images, or audio, based on training data. Prominent examples of GAI include GPT-4 and Midjourney, both of which are widely adopted, impacting work and communication practices [2]. The development of AI traces back to 1956 when John McCarthy coined the term “artificial intelligence.” In 1997, the supercomputer “Deep Blue” achieved a significant milestone by defeating the world champion chess player and establishing a clear benchmark in AI capabilities. More recently, in 2022, Midjourney, Inc. (San Francisco, CA, USA) released the Midjourney Discord bot. The refinement of large language models, such as ChatGPT, has been developed via various methods, including supervised learning, machine learning, and reinforcement learning. Moreover, latent text-to-image diffusion models such as Stable Diffusion and Midjourney have been widely applied in visual GAI software. Due to its capacity to process and analyze vast amounts of data quickly and accurately, AI is already being employed in diverse applications, employing natural language processing and autonomous vehicles. Its prevalence is expected to continue soaring in the coming years [3]. Machine learning is one of the key techniques used in AI. By learning from patterns and features in previous data, AI combines vast amounts of information with fast and iterative processing and intelligent algorithms to generate new content. In machine learning, AI is trained to identify patterns and make predictions based on existing data. By automating tasks that are currently handled by humans, AI allows for more time and resources for other activities that require human attention. In other words, its potential to enhance efficiency and productivity leads to cost savings, quicker delivery times, and improved overall performance [3]. This presents both opportunities and challenges across a wide range of professions and research areas, including education [4], healthcare [5], marketing [6], and transportation [7]. The application of GAI has been extensively explored in various design research, such as architectural design [8], fashion design [9], household appliance design [10], and industrial design [11]. In particular, Taiwan has gained recognition for its advanced footwear manufacturing technology and thriving footwear industry cluster.

This study aims to delve into the utilization of GAI to harness the accumulated industrial and technological knowledge and its influence on practical footwear design and the innovation process and investigate how GAI effectively integrates the industrial innovation process for footwear design. To examine the potential advantages and disadvantages of such integration, the creative concept development process of footwear design was incorporated into Midjourney V4, a text-to-image GAI. Midjourney V4 was selected based on its specific ability to generate visual outputs from textual input. Furthermore, to gain valuable insights into the impact of integrating GAI into the design process, seven experts in footwear design were invited to evaluate the results from the incorporation of Midjourney V4. The evaluation focused on assessing the creativity and feasibility of the generated designs. By involving the experts in the assessment, a comprehensive understanding of the implications and potential benefits of applying text-to-image GAI was obtained in footwear design. In this study, the various advantages that text-to-image GAI offers were explored in footwear design. Additionally, the influence of GAI on the overall design thinking process was assessed with a particular emphasis on creative concept development. By investigating these aspects, the result of this study suggests novel possibilities and challenges in the application of GAI in the footwear industry.

2. Literature Review

2.1. Defamiliarization

In the research of GAI, designers have challenges of the authority of a technocratic elite by employing the technique of defamiliarization. This approach allows for the present design or artistic works that highlight the disparities between machinic decision-making and human intuition [12]. The concept of defamiliarization was introduced by Shklovsky, who suggested that artistic or literary works can portray familiar objects or situations in an unfamiliar manner, thus prolonging the perceptive process and offering a fresh perspective [13]. Shklovsky’s examples were derived from literature showing how Tolstoy depicted ordinary things by describing them in intricate detail rather than using conventional names or signifiers. By avoiding precise nouns or conventional representations, these objects were presented as something distinct from well-known practices, prompting readers to think inductively rather than simply matching ideas with conventional concepts and archetypal images [14]. In particular, art removes objects from automatic perception to estrange common objects and present them in an unfamiliar light. This process slows down the reader’s perceptual experience and deautomates their perception and restoring awareness of perpetual change [15].

2.2. Design Thinking and Black Box

Design solutions are proposed to address specific problems. In the exploration of potential solutions, creativity is consistently recognized as a crucial element. However, design experts tend to rely heavily on their intuition during the generation phase [16]. This intuitive and enigmatic aspect of the design process often leads to a misconception of design expertise [17]. As a result, significant research has been devoted to demystifying creative design [16]. The black box metaphor is frequently employed to illustrate the process of creative development because the inputs and outputs of the design process are sometimes unobservable. Furthermore, the black box metaphor implies that failures in the design and creativity process remain hidden [17]. By providing clear descriptions of design methods and processes, it becomes possible to unveil designers’ private thinking and clarify the overall process. This enables other stakeholders to participate more consciously and rationally. These characteristics have a profound influence on designers, allowing them to adapt their working methods and thinking styles, with the Double Diamond model serving as a valuable reference [18].

The Double Diamond model, introduced by the British Design Council, represents a design process that incorporates both divergent and convergent thinking and forms cycles in an iterative development process. Divergent thinking enables designers to explore an issue more extensively or deeply, while convergent thinking involves taking focused action. The Double Diamond model consists of four distinct phases that shape the iterative process, namely Discover, Define, Develop, and Deliver [19,20]. The first phase, Discover, marks the initial stage of divergent thinking in the Double Diamond model. It allows designers to deeply understand the problem by engaging with stakeholders affected by the issues. The insights gathered from the discovery phase then assist designers in defining the challenge from a fresh perspective, completing the first diamond. Moving on to the second diamond, the Development phase encourages designers to seek inspiration and explore diverse possibilities. In this phase, different solutions and ideas are generated. Lastly, the Delivery phase prototypes and tests various solutions for designers to identify and reject those that are not suitable or effective [19]. Experienced designers often apply additional constraints during the solution generation phase to refine the solution and facilitate the generation of viable concepts. Throughout such a design process, designers modify goals and adjust constraints for a deeper understanding of the problem and progress in defining the solution. Despite these changes, designers endeavor to maintain their major solution concept for as long as possible. The purpose of these adjustments is to overcome challenges that emerge during the design process [16].

3. Method

3.1. Concept Ideation with Midjourney

The impact of GAI was explored on the application of both conventional and unfamiliar categories in the image generation process using Midjourney. Specifically, it investigated how GAI influenced the generation of different types of footwear designs. It was assumed that the existing subcategories of footwear were considered as known archetypal images while defining unfamiliar shoe styles, such as “protective devices for the feet”, which required a careful redefinition of component attributes. A lack of archetypal definitions increases conceptual creativity and flexibility but entails greater risks. On the other hand, explicit subcategory definitions may reduce the risk in design thinking but potentially limit creativity and innovativeness. To test these hypotheses, seventeen sets of footwear design proposals were generated using prompts with obviously different word counts and image styles. After decreasing homogeneity and similarity, ten sets of outcomes were selected for expert evaluation.

Table 1. Ten sets of prompts for Midjourney V4.

No.	Subcategories (Image Style)	Word Count of Prompts
1	Soccer shoe (Chelsea style)	10
2	Women’s sporty sneaker (Futuristic style)	15
3	Football shoe (Mutation style)	19
4	Police tactical boots (Geometric Style)	38
5	Running shoe (Geometric Style)	73
6	Lady running shoes (Futuristic style)	34
7	Children’s shoes (Vivian Westwood style)	59
8	Children’s rain boots (Vivian Westwood style)	54
9	High-heeled shoes (Japanese kimono-style)	45
10	Protective devices for the feet (No style)	220

provides a summary of ten prompts for Midjourney V4. Then, all ten sets of results generated by Midjourney are shown in Figure 1, Figure 2, Figure 3, Figure 4 and Figure 5.

3.2. Expert Evaluation

Seven experts in footwear design evaluated and examined the ten sets of conceptual designs using explanations in English and their Chinese translations. The 11-point Likert scale (ranging from 0 to 10, where 0 represented the least and 10 represented the most) was applied to subjectively assess the results in terms of prompt implementation, design creativity, and practical feasibility. Additionally, these ten experts described three aspects of the outcomes qualitatively to articulate professional insights.

Table 2. Profile of participating design experts.

No.	Gender	Age	Years of Experience
E1	Woman	34	10
E2	Woman	25	1
E3	Woman	37	12
E4	Man	35	6
E5	Woman	30	7.5
E6	Woman	23	1
E7	Woman	27	1

presents the profile of the design experts who participated in the evaluation process.

4. Results

4.1. Quantitative Analysis

In the descriptive statistical analysis, the sixth and ninth sets of generative results received higher scores in all three aspects. The tenth set received the lowest scores in prompt implementation. The seventh set scored the lowest in creativity, and the third set scored the lowest in feasibility (Figure 6). The results of ANOVA indicated significant differences in prompt implementation (F [9, 69] = 2.165, p = 0.037), creativity (F [9, 69] = 4.017, p = 0.000), and feasibility (F [9, 69] = 3.598, p = 0.001). The post hoc test result (Duncan) for prompt implementation revealed a significant difference, with the 9th set (7.86) scoring significantly higher than the 3rd set (5.29), 7th set (5.00), 4th set (4.71), 5th set (4.71), and 10th set (4.57). In creativity, the post hoc test (Duncan) indicated that both the 3rd set (7.71) and 9th set (7.71) scored significantly higher than the 4th set (4.43) and 7th set (3.71). For feasibility, the 8th set (8.00), 6th set (7.43), and 4th set (7.29) scored significantly higher than the 10th set (3.86) and 3rd set (3.57).

4.2. Qualitative Analysis

Qualitative assessments were conducted for prompt implementation, design creativity, and practical feasibility by the experts. The summary of the qualitative assessment regarding the implementation of prompts is presented in

Table 3. Key assessment for prompt implementation.

Expert	Ideation	Comments
E1	9th	I found this set of results to be highly artistic, with almost all the prompts successfully fulfilled, except for the generation of fish scales.
E4	9th	The Japanese style is evident in the presentation.
E5	9th	The implementation rate of prompts is not high, and many conditions interfere with each other.
E3	10th	The implementation rate of prompts is not high, and many conditions interfere with each other.
E5	10th	The settings and conditions in the prompts are specific and detailed, but GAI lacks integration capability, essentially only fulfilling the first two prompts.

,

Table 4. Key assessment for creativity.

Expert	Ideation	Comments
E1	9th	GAI doesn’t excel as much in the creativity of footwear design, perhaps because it analyzes big data using the shoe’s original stereotypes and archetypes to determine the shape the shoe should be.
E3	9th	GAI can combine difficult-to-imagine style features to generate designs rapidly.
E2	3rd	I find these computation results very innovative and creative! However, the design of this footwear makes it challenging to convey the elegance of women’s high-top shoes.
E2	4th	In terms of creativity, I believe these results are rather conservative, resembling the boots available in the market. Therefore, from a development standpoint, they seem quite reasonable.
E6	4th	These are relatively common among AI-generated results, but I believe they would be more widely accepted. Moreover, the development process should be easier compared to other results.
E4	7th	The prompts are more specific, resulting in a lower level of creativity.

and

Table 5. Key assessment for feasibility.

Expert	Ideation	Comments
E6	8th	I personally find it quite attractive, and the development process shouldn’t be difficult either. This design has a certain charm that makes you want to buy it and keep it as a collectible.
E4	8th	The shiny polyethylene and transparent materials are presented reasonably.
E2	6th	Most sock-style sneakers tend to enhance the stability of the heel. In this set of results, I noticed that both the upper and the heel counter are supported by TPU or leather materials.
E5	10th	Regarding setting prompts, I recommend using no more than three design conditions for design generation and then using them as inspiration for the design.
E1	3rd	It has a visual impact but lacks consideration for wearability.
E4	3rd	It is challenging for GAI to simultaneously assess the practical feasibility of both materials and craftsmanship in the development process.

.

5. Conclusions

The application of GAI in footwear design was evaluated from three perspectives. The results of the quantitative analysis showed that footwear designs referencing particular cultural styles demonstrated creativity and prompt implementation, especially in the case of the ninth set of results. However, deliberate avoidance of the specification of subcategories for footwear and refraining from incorporating specific cultural styles hindered the high integration of design definitions as suggested by prompts. In design creativity, the results exhibited a negative correlation between creativity and feasibility. For instance, the fourth, seventh, and eighth sets lacked creativity due to their excessive feasibility and similarity to existing products. Moreover, the majority of low feasibility resulted from shattered designs or sharp features that conflicted with comfort requirements, resulting in relatively low feasibility. Expert evaluations also indicated that GAI lacked integration, particularly when design conditions interfered with each other. Additionally, GAI’s prediction of material wearability during usage still requires improvement. These are areas where GAI can potentially enhance its design applications in the future.